Interactive Demos
Iterative Timestamped Audio Describer
1 · The Describer: iterative structured perception
Multi-pass describing
One clip, built up pass by pass
Open demo
2 · Understanding: describe, then reason
MMSU · MSU-Bench · Daily-Omni · Video-Holmes
Describe-then-reason across four benchmarks
Open demo
3 · Generation: description → controllable audio
Word-level emphasis
Emphasis follows the marker
Open demo
Reconstruction
Real → description → re-generated
Open demo